Normalise a Histogram excluding NaN?

Shown under M1, i am attempting to plot a histogram and it's Normal Distribution, however the data has a large quantity of NaN. I was able to omit them in the plotting for the line but not the histogram bars themselves.

Does anyone know how to fix this?

Answer 1

[Star Strider]

One option is:

histogram(m(~isnan(m)), 50, 'Normalization','PDF')

This eliminates the NaN values from your data before doing the analysis.

Experiment to get the result you want.

Answer 2

[Erik Giesen Loo]

Using a deprecated method, but may be useful to compare the validity of MATLAB's histogram. I start by defining a function that finds the length of a vector ignoring all nan's (similar to how nanmean and nanstd work).

function nx = nanlength(x)
% nanlength(x) returns the number of elements of vector x that are not nan.

dim = find(size(x) ~= 1, 1);
if dim == 2
  x = x(:)';
else % dim == 1
  x = x(:);
end

xnans = isnan(x);
if any(xnans(:))
  nx = sum(~xnans,dim);
else
  nx = size(x,dim); % a scalar, => a scalar call to tinv
end

Now we proceed to implement the pdf normalization ourselves:

[y,x] = hist(data,m); % m = number of bins using any desired algorithm
width = x(2) - x(1);
y_bar = y/(nanlength(data)*width);
bar(x,y_bar,'hist')

It might be nice if MATLAB implemented an option to ignore NAN's like in many other functions.